Performance Comparison of the Cogent Confabulation Classifier with Other Commonly Used Supervised Machine Learning Algorithms for Bathing Water Quality Assessment

Abstract

The purpose of this study was to implement a reliable model for bathing water quality prediction using the Cogent Confabulation classifier and to compare it with other well-known classifiers. This study is a continuation of a previously published work and focuses on the areas of Kaštela Bay and the Brač Channel, located in the Republic of Croatia. The Cogent Confabulation classifier is a thorough and simple method for data classification based on the cogency measure for observed classes. To implement the model, we used data sets constructed of remote sensing data (band values) and in situ measurements presenting ground-truth bathing water quality. Satellite data was retrieved from the Sentinel-3 OLCI satellite and it was atmospherically corrected based on the characteristics and specifications of band wavelengths. The results showed that the Random Forest, K-Nearest Neighbour, and Decision Tree classifiers outperformed the Cogent Confabulation classifier. However, results showed that the Cogent Confabulation classifier achieved better results compared to classifiers based on Bayesian theory. Additionally, a qualitative analysis of the four best classifiers was conducted using spatial maps created in the QGIS tool.

A. Ivanda, L. Šerić, M. Braović, and D. Stipaničev, “Performance Comparison of the Cogent Confabulation Classifier with Other Commonly Used Supervised Machine Learning Algorithms for Bathing Water Quality Assessment,” CIT. Journal of Computing and Information Technology, vol. 30, no. 1, pp. 1–21, 2022, doi: 10.20532/cit.2022.1005436.